Codex Cloud Tasks: How to Start and Plan Requirements

A Codex cloud task keeps running even when you close your laptop or let it sleep. In ChatGPT, choose "Work in > Cloud", connect a repository, then hand off the task (checked Sept 30, 2026, per the official docs). The docs do not state the announcement date or the start date. Plans from Plus ($20 per month) up list web, CLI, IDE and iOS use in the table, and cloud tasks share the same plan limit as local messages.

Three lines to read first

• In its DevDay 2026 announcement, OpenAI said "Codex Cloud runs tasks while your laptop is closed, with access across devices".
• Each task gets its own workspace, and you can pick up the results on web, mobile and the desktop app.
• There is no separate cloud-only limit. Cloud tasks draw from the same pool as messages you send locally.

Codex cloud tasks work differently from local Codex

A cloud task reads, edits and tests code in a workspace on OpenAI's side, not on your computer. Local Codex needs your laptop to stay on, and it stops when you close the lid. Cloud has no such limit. The official docs say: "Each task has its own workspace and can keep working while your computer is asleep".

Here is a hypothetical example. Just before leaving work, you ask "Find why the payment module tests fail and fix it", then close your laptop. This is an illustration of how you might use it, not an actual run result. The next morning you review the changes and test results. If you like them, you commit or open a pull request (PR, a proposal to merge code changes). The docs give no numeric time limit for a task. McPhee's view is that for long tasks, it is better to check an interim result once.

How to use Codex cloud tasks in 3 steps

The official Codex Cloud docs (learn.chatgpt.com/docs/cloud) give this order.

1. Open ChatGPT and choose "Work in > Cloud" as the work location.
2. Create an environment. When you select a repository, Codex looks over the project and runs the setup steps. It asks for any missing information or permissions. If a GitHub connection window appears, approve it.
3. Choose the environment, click "Start a new task" and write your request.

When the task ends, review the changes and test results. Then send a follow-up request, or hand off to a commit or PR. In the network access settings, you can allow the internet or allow only specific domains. For projects that must download external packages, this setting often blocks the install step.

Which plan do you need?

The plan table in the official Codex pricing docs (learn.chatgpt.com/docs/pricing) is below. I did not find a table that lists cloud task availability in a separate column for each plan. The "supported environments" column in the table is the basis.

Plan	Monthly price	Listed in the table
Free	$0	GPT-6 Luna default speed, desktop app
Go	$8	GPT-6 Luna default speed, desktop app
Plus	$20	Web, CLI, IDE, iOS, GPT-6 Sol/Luna, credits can be purchased
Pro	$100 to $500	Includes Plus features, Astra Ultrafast only on the $500 plan
Business	$20 per user (annual billing, $25 with monthly billing)	Desktop and mobile, larger VMs, credit purchase, SAML SSO
Enterprise/Edu	Separate contract	Business features plus SCIM, RBAC, audit logs

In the table, Free and Go list only the desktop app, with no web, CLI or IDE. McPhee's view is that it is safer to plan on Plus or higher for cloud tasks. This is an inference, because the docs do not say it outright. Before you pay, check whether the "Cloud" option opens in the Codex screen of your own account.

Local and cloud share the same limit

The official Codex pricing docs say local messages and cloud tasks share the same plan limit. So if you use many local messages during the day, a cloud task you hand off at night can hit the limit. The limit is based on a 5-hour window.

Model	Plus, local messages per 5 hours
GPT-6 Astra	5~45
GPT-6.1 Sol	15~160
GPT-6 Sol	15~150
GPT-6 Luna	350~3,000

The range is wide because the weight of one message depends on the size of the task. A task that reads a whole large repository uses far more than a short question. Standard Business lists the same numbers in the table. Pro is marked as having no 5-hour limit, but policy changes often, so check the screen before you pay.

If you go over the limit, you buy credits to keep going. Credit rates per 1 million tokens for GPT-6.1 Sol are 50 for input, 2.5 for cached input and 250 for output. Astra is 5 times more expensive, at 250 for input and 1,250 for output. A fuller price comparison is in GPT-6.1 Sol price and Astra comparison.

What to check when using it in Korea

• Billing amount: The prices above are in US dollars. The exact charge in Korean won (KRW) and VAT show only on the checkout screen. McPhee does not give a KRW conversion without an official source.
• GitHub account: Cloud tasks require a connected repository. Company organization repositories often need admin approval.
• Company code: You are effectively uploading the repository to a cloud workspace, so first check whether your internal security policy allows external transfer.
• Time zone: The 5-hour window runs regardless of Korea time. A morning work session and an evening handoff can fall in the same window.

Common mistakes

• Thinking cloud tasks are separately unlimited. They share the limit with local.
• Thinking local Codex keeps running when you close the laptop. Only tasks started in Cloud keep running.
• Handing off a task that needs package installs while internet access is blocked. Set the network settings first.
• Merging the PR without checking the result. Decide after looking at the test results and the list of changed files.

Frequently asked questions

Can I run Codex cloud tasks for free?

In the plan table, Free and Go list only the desktop app. The docs do not separately say whether cloud tasks are available, so check in your account screen.

How long can a task run?

The docs give no numeric time limit. The honest answer is that it is unknown.

Can I view it on my phone too?

The docs say you can review and continue tasks on web, mobile and the desktop app.

Official sources

• OpenAI DevDay 2026 announcements and developer resources
• Official Codex Cloud docs
• Official Codex pricing and usage docs

McPhee left out items the docs do not cover, such as the cloud task time limit and per-plan cloud availability. Start your first task with one small bug, and check the price screen once more before you pay.